Global deforestation patterns: comparing recent and past forest loss processes through a spatially explicit analysis/Tendances de la deforestation au niveau mondial: comparaison des processus passes et recents a travers une etude spatialement explicite/Tendencias globales de deforestacion: comparacion de procesos recientes e historicos de perdida de cubierta forestal a traves del analisis explicito de datos espaciales.
A good understanding of forest conversion to other land uses is important for the development of policies and measures aiming to reduce the loss of forests and its associated carbon emission. Current day deforestation concentrates in the tropics (FAO 2016). A global forest conversion analysis based on Landsat samples (FAO and JRC 2012) quantifies tropical deforestation at 6.8 million ha/year between 1990-2005, or nearly 2.5 times the net forest area gained in the other climatic domains combined (boreal, subtropical and temperate).
Deforestation rates by climatic domain have changed dramatically over the past centuries. Williams (2006) estimates that deforestation in the temperate zone outpaced deforestation in the tropical zone for over two centuries between 1700-1920. After that deforestation rates dropped steeply in the temperate region where reversion to forest happened at a large scale (Richards 1990 in Williams 2006). To the contrary, population growth and slowly growing economies in the tropics increased deforestation rates tremendously, increasing with a factor >7 in the period 1920-1950 compared to the preceding 70 year period (Richards 1990 in Williams 2006).
Forest loss is the result of many processes driven by multiple causes. Deforestation is human induced forest loss. We can distinguish direct causes and underlying causes of land conversion, referred to as 'drivers and pressures' by Smith et al. (2010) and 'underlying and proximate causes' by Geist and Lambin (2001) respectively. Underlying causes (e.g. societal trends, socio-economic, technology and climate factors) determine the degree of direct causes (e.g. conversion to cropland or infrastructure, conversion due to severe natural disturbance) resulting in land use change.
A quantification of drivers of deforestation by geographical region is given by Hosonouma et al. (2012) who used information reported in REDD+ Readiness Preparation Proposals and the Global Forest Resource Assessment (FAO 2010). The authors suggest agriculture (cropland and pasture) to be by far the largest direct cause of deforestation; according to their estimations between 70-80% of forest conversion is to agriculture in Africa, around 70% in (sub)tropical Asia and >90% in Latin America. Other studies equally point at agricultural expansion to be the largest direct cause of deforestation in Africa, Asia and Latin America (Nepstad et al. 2009, Gibbs et al. 2010, Guitierrez-Velez et al. 2011). Williams (2006) sets out how deforestation processes changed over time, from indigenous smallholders, to colonial plantations and commercial clearings, but all processes point at one common direct cause of deforestation in the past: agricultural expansion.
This study examines how deforestation relates to a selection of biophysical and socio-economic land characteristics that were expected to approximate or regulate direct and underlying causes of deforestation, notably agricultural expansion hypothesized to be the main direct cause of deforestation. It compares these relations for historical and recent deforestation, hypothesizing that past and recent deforestation patterns have similar empirical relations to the tested variables. If these relations show similar patterns for past and present deforestation, this indicates the variables examined could hold predictive power in identifying locations of likely future change. The use of this predictive power is illustrated creating a deforestation risk map for Africa and overlaying this with protected areas to understand the different pressures on protected forests on the continent. The study finally plots recent deforestation rates against economic development (approximated with GDP value and increase rate), hypothesizing the deforestation rates to show similar dynamics as those predicted by the forest transition theory.
To examine whether a meaningful pattern would be found when comparing the locations of historical and recent deforestation with variable values of those locations we took the following steps: (1) select socio-economic and biophysical land characteristics (hereafter referred to as variables) which we expected to have an influence on deforestation and which are available in map format with global coverage and a meaningful resolution (Table 1); (2) create a map with an approximation of historical (pre-agriculture) forest cover and create a map of actual forest cover where the comparison of both provides us with locations where forest has been lost over time; (3) assess locations of recent deforestation using detailed sample information of a global remote sensing survey which assessed forest land-use change for the period 1990-2005; (4) explore whether historical deforestation concentrates on locations with certain variable values and whether recent deforestation concentrates on locations with certain variable values; (5) compare the patterns of historical and recent deforestation when plotted against the selected variables.
Additionally, as the expectation was that the variables were related to one another, correlations between the variables were explored. The analysis looking at correlations among the variables helps to understand whether patterns of deforestation were a result of a common element in the variables or whether the variables interact independently of one another with deforestation.
Finally, those variables which in the above steps reveal strong explanatory power for both historical and contemporary forest loss are used to create a deforestation risk map for Africa which is overlaid with protected areas to understand the different pressures on protected forest in the continent. This is done to illustrate how these correlations can be used on a more local scale for different analytical and planning purposes.
In the following sections the above described steps are explained in further detail.
Selection of socio-economic and biophysical land characteristic maps
The variables were selected because they had an expected relation to direct or underlying causes of deforestation. Most of these are "common sense" factors (e.g. agriculture expansion is more likely to happen on forests located on accessible land with high crop suitability). The variables are described in Table 1. Table 1 does not include variables such as slope, soil type or elevation since these are already included in the accessibility and crop suitability map.
Approximating historical deforestation
A map providing a proxy for historical forest extent was constructed by identifying locations where biophysical conditions favour forest in the absence of human induced or natural disturbances using ecozones defined in FAO (2012). Historical forest extent in this study is an approximation of pre-agriculture forest which assumes to go back roughly 6000 years in time when--to our best available knowledgesignificant human induced forest clearing started by the first sedimentary farmers in Europe (Williams 2000). However, the map provides a reflection of current day climatic conditions and associated vegetation types, therefore forest loss over this extensive period due to climate change is not included. Available historical forest maps were not used since they only inventoried dense or hardwood forests which do not compare with the current 10% coverage in the FAO definition. For example, a forest assessment dating beginning of last century estimated a total area of 3 billion ha (Zon and Sparhawk 1923) which is about 1 billion ha below the forest area assessed for the year 2000 (FAO 2010). It is usually assumed that global forest area has been shrinking and most of this large discrepancy between these assessments is thought to be due to the changing definition (Mather 2005).
The global ecological zones map is an approximate equivalent of Koppen-Trewartha climatic types (Trewartha 1968), in combination with vegetation physiognomy and one orographic zone. From the ecozones described in Table 2 all zones with 'forest' in their names were assumed to be fully forested plus part of the mountain systems. Unlike the other ecozones, mountain systems in FAO (2012) are not associated with one predominant vegetation type as they are approximated by elevation. Areas with biophysical potential for forest in mountain systems were approximated using the static natural vegetation model BIOME (PBL 2013) defined by soil types and climatic conditions (Prentice et al. 1992). The proxy for historical forest extent is shown in Figure 1 (representing the sum of the 2000 forest area and historical deforestation proxy classes).
To create a historical deforestation map we overlaid the historical forest cover map with the forest map for the year 2000 assuming that all non-forest in this comparison provides an estimate of historical deforestation (Figure 1). The forest map for the year 2000 was only used to assess historical deforestation, since for the location analysis comparing recent deforestation with the variables a more accurate sample based forest change analysis was used as described below. The 2000 forest map was created as follows. The world was stratified in climatic domain and administrative region combination resulting in 19 strata (Table 3). The forest area quantity in each stratum is assessed though a Landsat sample-based analysis of forest/non-forest for the year 2000 described in FAO and JRC (2012). The forest area in each stratum is determined using the Horvitz-Thompson direct estimator following Eva et al. (2010)--that is, by calculating the mean proportion of forest over all sample sites within the stratum and multiplying this figure by the total land area in the stratum. Accordingly, a mosaic of the MODIS Vegetation Continuous Fields (VCF) dataset for the year 2000 (DiMiceli 2011) was created using bilinear resampling to aggregate the MODIS data from the original 250x250m to a 10x10km spatial resolution. Next we allocated the 2000 forest quantity per stratum from the Landsat sample-based analysis to the highest VCF values in our newly created 10x10km MODIS VCF map until the 2000 forest quantity was met. The resulting VCF thresholds per stratum in the 10x10km MODIS VCF map are listed in Table 3. The 10x10 km resolution wall-to-wall map of forest extent in the year 2000 is displayed in Figure 1.
Assessing recent deforestation
Recent deforestation was assessed through a Landsat sample-based forest conversion analysis for 1990-2005 as described in FAO and JRC (2012). Given the concentration of deforestation in the tropics, only the results from the tropical domain are included. To exclude all samples falling in the desert or other locations without forest, only those samples having forest in the year 1990 are included resulting in a final number of 2522 samples used in the analysis comparing locations of recent deforestation with corresponding variable values. The detailed deforestation location information inside each 10 by 10km sample (FAO and JRC 2012) is for the analysis aggregated to a 10 by 10 km resolution.
Spatial analysis to explore correlations with historical deforestation
The historical and recent deforestation assessments are compared with spatially explicit assessments of a selection of variables (Table 1) to analyse whether deforestation shows a meaningful relation to these variables. The variable maps were resampled to match the pixel resolution of the historical forest/non-forest map (10 x 10 km). An overlay with the variable maps was performed using Idrisi's crosstab function and assigning classes to the continuous variable maps. Accordingly, the percent deforestation value of each class in the variable map was computed as the ratio of the quantity of non-forest grid cells in that class to the quantity of all grid cells in that class (Figure 2).
To assess the strength of the relation between the variable maps and past deforestation we use the Relative Operating Characteristic (ROC). ROC is a method to compare a rank variable, such as Cost Distance, to a Boolean variable, such as the existence of non-forest in 2000 (Pontius and Batchu 2003). The ROC procedure analyses each variable at multiple thresholds to measure the density of non-forest between each pair of sequential thresholds. Thus the procedure can detect the ranges within the variable that are either positively associated or negatively associated with non-forest. The area under the ROC curve (AUC) is a summary measure of the strength of the relationship. An AUC of 1 means a perfectly strong positive relationship, AUC of 0 means a perfectly strong negative relationship, and an AUC of 0.5 means a perfectly random relationship.
Spatial analysis to explore correlations with recent deforestation
The 2522 sample centre points of the 1990-2005 deforestation assessment are projected on the variable maps to collect the variable data at the centre point. For each sample point, the values of the four nearest neighbouring pixels of the variable map were averaged and the average value was attributed to the point. Therefore one sample will be allocated a single variable value. If no data was available for the specific location of the centre point those samples were not considered in the analysis. Accordingly we group the continuous variable information in classes and collect the average relative deforestation in this class with a 90% confidence interval which we plot on graphs. A Mann-Whitney U test was used to analyse whether the sample values in the different classes were significantly different.
Correlation among variables
Kendall's tau-b correlation coefficient was used to measure the association between the variables. Kendall's tau-b is a nonparametric measure of correlation coefficient used for variables measured on an ordinal scale and has fewer statistical assumptions the data must comply with (IBM 2011). It is an alternate to the commonly used Pearson's correlation coefficient but does not require the data to meet statistical assumptions of normality and linearity. When Kendall's tau-b is 0, it indicates the independence of the variables. The maximum value for Kendall's tau-b is 1, indicating a perfect positive association between the variables and the minimum value is -1, indicating a perfect negative association. In order to reduce spatial autocorrelation and data size, historically deforested areas are randomly sampled (n=5845 10x10km pixels). Each pixel is an observation for the variable by comparing the values of each of the variables at each pixel we construct a Kendall's tau-b correlation matrix. Kendall's tau-b was calculated using SPSS statistical software.
Exploring the relation between recent deforestation and economic development
Economic development is arguably the most important underlying process of the forest transition theory. This theory predicts a systematic pattern of forest cover change in a country over time. In the first phase, a country has a high and relatively stable portion of land under forest. Next deforestation begins and accordingly accelerates, reaching high levels, after which it slows down again. Chomitz et al. (2007) and Angelsen and Rudel (2013) provide some economic, social and demographic characteristics to describe the phases of the forest transition theory. We approximate the socio-economic development underlying the forest transition curve by grouping countries into income and income change classes as follows: pre-transition is approximated as low income countries with low income change, early transition is approximated as low income countries with high change, late transition is approximated as middle income countries, and post-transition is approximated as high income countries (see Table 4). Deforestation as percentage of forest in the samples is accordingly plotted in the different transition phases.
Deforestation risk in Africa's protected forests
To illustrate the application value in case of a correlation with the tested variables, an analysis was undertaken to assess the deforestation pressure on protected forests in Africa. For this purpose, a deforestation risk map is created and the map is overlaid with protected areas. To create a deforestation risk map, first a 2010 wall to wall forest cover map for Africa is created. The forest cover map was created using MODIS VCF 2010 percent tree cover data, the global ecological zones (GEZ) and forest area data from the Global Remote Sensing Survey (FAO and JRC 2012). The VCF data was downloaded from http://www.landcover.org/data/vcf/ at 250m resolution in MODIS sinusoidal projection and the GEZ data was downloaded from the FAO GeoNetwork. The pre-processing procedures involved creating a mosaic of the VCF tiles and aggregating the data to a 10km by 10km spatial resolution. Forest area for each ecological zone was determined using data from the Global Remote Sensing Survey (FAO and JRC 2012). The forest area quantity in each stratum is assessed though a Landsat sample-based analysis of forest/non-forest for the years 2000 and 2005. The trend from 2000 to 2005 is extrapolated to 2010 to calculate the forest area in 2010. The forest map was created by allocating the quantities of forest by ecological zone and region to pixels with the highest MODIS VCF percent tree cover values. With 2010 forest cover established, historically deforested areas were identified as locations with the biophysical potential for forest cover in the absence of anthropogenic and natural disturbances.
GEOMOD (Pontius and Chen, 2006) was used to determine the "deforestation risk" of the forest in the resulting forest/non-forest map, where the variables accessibility, rural population density and crop suitability were used to assess this risk. The deforestation risk is an approximation of the pressure these forests are under to be converted, as under perfect law enforcement the actual risk would be zero in the protected areas.
The location of forests with the primary function of conservation were determined using the World Database on Protected Areas (WDPA). WDPA gives spatial and attribute information on over 190,000 nationally and internationally protected sites at a global scale. Protected areas in Africa with a designated status and classified as IUCN category I-IV are included in the spatial analysis of protected areas.
RESULTS AND DISCUSSION
Historical and recent deforestation
Table 5 displays the results of the approximation of historical deforestation and the results of the assessment of recent (gross) deforestation from the global forest land-use assessment (FAO and JRC 2012). The results show that recent deforestation concentrates in the tropics. This finding is confirmed by a recent global tree cover change study (Hansen et al. 2013). However, the approximation of historical deforestation indicates this situation to have been quite different in the past. Table 5 reveals that when considering the accumulated net deforestation over centuries, losses (in %) in the temperate and tropical domain are comparable. This is in agreement with the narrative set out by Williams (2006) stating tropical deforestation was preceded by centuries of deforestation in the temperate zone. Table 5 should be interpreted with some care given the rough approximation of historical forest cover. Some ecological zones like dry forest are likely to have natural occurrence of fewer than 10% tree canopy coverage without this being due to deforestation or degradation. Therefore, Table 5 displays a range for both the tropical and subtropical domains providing an estimate including and excluding dry forest from the approximation of historical forest cover. In the remainder of the analysis, dry forests are included in the proxy of historical forest area. Given that in many tropical transition zones deforestation is expected to be overestimated in Figure 1, historical net deforestation in the temperate domain may be larger than net deforestation in the tropical domain.
Relating variables to historical deforestation
The graphs plotting the historical deforestation (%) per variable class are displayed in Figure 3-7a. The results of the ROC analysis give an indication of the strength of the relation between historical deforestation and the variables (Table 6). Ruminant livestock density is not included in the ROC analysis in Table 6 due to the low amount of variable classes available in the map. This limits the number of thresholds provided by the ROC curve which can skew the AUC value. Rural population density has the highest AUC of the variables, followed by accessibility and crop suitability respectively. These results are in line with expectations when assuming agricultural expansion to be the largest direct driver of deforestation.
Accessibility and historical deforestation are negatively correlated (deforestation happens more in locations with at lower travel time) while rural population, crop suitability and ruminant livestock density are positively correlated with historical deforestation (e.g. deforestation happens more in locations with more rural people per km2). These global findings for the tropical domain are similar to findings by Ernst et al. (2012) in the Congo Basin for population density and accessibility. A combined interpretation of Figure 3-7 and Table 6 reveals monotonic relationships for all the variables except pasture suitability. The pasture suitability graph (Fig 7a) displays a random pattern and the AUC value (Table 6) is close to perfectly random, thus the pasture suitability map does not seem to exhibit explanatory power for the patterns of historical deforestation. The lack of correlation with pasture suitability yet the correlation found between deforestation and livestock density seems to indicate pasturing can be implemented in a variety of landscapes with highly varying stock densities and productivity.
Comparing historical and recent deforestation patterns
Past and recent deforestation relate to rural population density and accessibility in similar ways. For recent deforestation the higher rates of deforestation are in areas with shorter travel times and higher population densities--which is highly intuitive. There is some variability in the shortest travel time areas with the highest population densities--which may reflect locations near cities where deforestation rates are slowing and possibly forestation could be happening to satisfy demands for forest services and products. Reducing pressures on forest resources as associated with urbanization are suggested by Lambin and Meyfroidt (2010) and Rudel et al. (2005). In a comparable study by Mayaux et al. (2013) in the African rainforest a decrease in deforestation in the highest population density class is seen strengthening the hypothesis of declining deforestation rates in the vicinity of cities.
Land with high crop suitability seems to be more targeted for conversion as would be expected, but no significant difference is seen between deforestation on good and moderate crop suitable land. This may be explained by technologies capable of ameliorating previously unsuitable terrain to allow profitable crop production. For example, in some locations agriculture is expanding on land which according to the GAEZ is classified as not suitable thanks to the irrigation of deserts (Alexandratos and Bruinsma 2012). Another explanation could be that land with higher crop suitability which is at the same time highly accessible is either already in use or located in protected areas. We found that 24% of the remaining forest in the tropics is located on land with good to very high crop suitability, however less than 30% of this forest on good crop suitable land is located at a travel time of <6 hours (Table 7).
Correlations among variables
The correlation matrix (Table 8) shows no strong correlations between any of the variables. The highest positive correlation exists between rural population density and ruminant livestock density and the highest negative correlation exists between rural population density and accessibility. These relationships might indicate the majority of livestock concerns small-holders therefore livestock increases with rural population density. The proximity of markets could be an important determinant for the presence of rural activities and thus the rural population explaining the correlation between rural population density and accessibility.
The lack of strong correlation between the variables suggests that the correlations they display with deforestation are not duplications of a common element in the variable maps. Therefore these results suggest each variable that displays a correlation with deforestation may add complementary explanatory power to the occurrence of deforestation.
Underlying socio-economic processes
A major underlying cause of change is not adequately captured with the variables tested for their correlation with forest loss: economic development.
Figure 8 shows that forest loss changes with the different phases of economic development of a country in a manner which is consistent with the deforestation rates associated with the forest transition curve. This finding suggests that policies and measures aiming to reduce the loss of forest like REDD+ should take the different forest transition phases into account as suggested also by Angelsen and Rudel (2013), especially for countries in the pre-transition phase. It is likely that forest loss rates in pre-transition countries will increase as they move to the early and late transition phases while the forest loss rates of late transition countries are more likely to go down moving towards post-transition. Shandra et al. (2009) suggest that this differentiation in deforestation rate may be strengthened by rich nations externalizing their resource demands and environmental degradation onto the poor nations of the world through the vertical flow of exports. As such, average historical deforestation rates may be a poor predictor of the pressures poor countries face on their forests once their economy starts accelerating. Forest loss in the early and late transition phases are not significantly different showing little evidence of stagnating deforestation when economies move towards middle income GDPs.
We distinguish between forest loss and tree cover loss, in which forest refers to forest land use which is defined by the human activities and inputs on a given land area and tree cover corresponds to the observed biophysical properties of the land surface. As such temporarily destocked land for which the use remains forest is not considered forest loss. Forest loss comprises both forest land use conversion as a result of natural disturbance and human-induced forest loss (deforestation). We consider most forest loss to be humaninduced and as such the terms are used interchangeably in this article. When creating the 2000 wall-to-wall forest map there is disagreement when allocating forest areas assessed by the global remote sensing survey to VCF values. The assessment concerns the forest land use whereas VCF corresponds to tree cover. This allocation disagreement is thought to be greatest in landscapes where even-age forest management is practiced using clearcutting followed by either natural regeneration or replanting. While this is more common in the temperate zone it also occurs in the tropics--particularly in planted forests and shifting cultivation. However, it is expected that this analysis of change in the tropics is not significantly influenced by the differences between the VCF and Landsat-based analyses used in map preparation for this paper.
The creation of the proxy of historical deforestation map will contain errors in transitional ecozones where the natural vegetation is such that forest and non-forest are naturally scattered in a mosaic or patch like structure. The natural climax non-forest in these locations will erroneously be labelled as deforested in this approach (e.g. we expect this to happen especially in dry forest, wooded tundra and mountain systems). For this reason, a range of historical deforestation is provided in Table 5. This may explain the unrealistic 'peaks' in the correlation graphs, notably in the unsuitable classes of crop and pasture suitability in Figure 6a and 7a, which do not show in the recent deforestation analysis (Fig 6b and 7b). The higher deforestation in the not suitable classes therefore likely responds to misclassifications of biophysical hostile locations for trees where even in the natural climax condition tree cover is expected to be absent (e.g. a mountain slope or desert).
There is a 10-year overlap between the periods used to approximate historical and recent deforestation in this study (4000 B.C.-2000 A.D. and 1990-2005). This overlap could not be avoided since VCF data wasn't available for the year 1990. Deforestation between 1990-2000 comprised only 4% of the historical deforestation proxy which the authors therefore consider to be insignificant and for this reason we believe the historical and recent deforestation analyses can be considered independent.
Another limitation of the study is associated with the variables used most of which correspond to current day situations and do not (fully) capture dynamics which may have been associated with deforestation in the past.
Illustration of application: Deforestation pressure in Africa's protected areas
The deforestation risk map based on the predictive power of the variables accessibility, rural population density and crop suitability for 2010 forests in Africa is displayed in Figure 4.
The protected areas are overlaid on the deforestation probability map to identify the pressure on forests in- and outside the protected areas. The results of this analysis are displayed by region in Table 8. The regions are defined by the Global Administrative Unit Layers (FAO 2014).
Table 8 shows a systematically lower pressure on forests inside protected areas compared to the pressure outside protected areas for all regions. This is to be expected as governments have to consider the needs of their populations for agricultural expansion and as such may select less densely populated and more inaccessible forests for a protection status. That protection tends towards areas which are less likely to be cleared is also found by Joppa and Pfaff (2011). The difference in pressure on forest in- and outside protected areas is largest for Northern Africa followed by Eastern and Western Africa. Pressure on forests in protected areas is highest in Southern Africa though the total forest area is smallest in this region. Western and Eastern Africa are the next regions with highest pressure with a substantial forest area in Eastern Africa. The high pressure on West African protected forests is confirmed by the sharp boundaries of forest cover at their edges (Joppa et al. 2008). The pressure is relatively low in Central Africa where most forests are located. Finally, Southern and Easter Africa have the relatively largest areas of their forests with a protected status which may be a signal of depletion outside protected areas highlighting the importance of protected areas in halting deforestation. When interpreting these results, it should be noted though that this probability calculation does not consider trend breaks or changing pressures such as explained by the forest transition curve. These results indicate the efforts to conserve remaining forest may be very different in the regions compared.
Contemporary deforestation concentrates in the tropics. In comparing historical (4000 B.C.--2000 A.D.) and recent (1990-2005) deforestation, this study suggests though that accumulated net forest loss over centuries (in %) is comparable in the temperate and tropical domains. Deforestation has many layers of complexity with a variety of underlying causes that are difficult to generalize and that have changed over time. However, there is evidence for recurring patterns of deforestation and similar correlations are found with variables related to the direct causes of deforestation. At a global scale correlations were found between deforestation and costdistance, rural population density, crop suitability and ruminant livestock density. No correlation was found between deforestation and pasture suitability, indicating that pasture can be implemented on a wide range of landscapes. The fact that these recurring patterns are comparable for past and recent deforestation suggests the correlations to be empirical and therefore future deforestation is quite likely to follow a similar pattern. The predictive power of these variables is used to assess pressure on land for future deforestation in Africa in- and outside protected areas. This reveals protected status to concentrate in forests with lower than average deforestation pressure and it reveals the pressure on forests in protected areas to vary greatly per geographic region. Finally, recent deforestation rates are plotted against economic development revealing lower deforestation rates in poor countries with low GDP growth and in rich countries. This finding indicates that poor countries are likely to face large pressure on their forests when economic development accelerates.
We recommend the use of variable maps in spatially explicit models to identify locations with higher probability of future deforestation (e.g. Arekhi 2011, Rosa et al. 2013, Sonter et al. 2014, Wassenaar et al. 2007). When such deforestation probability maps are prepared at a national scale using more detailed available data, they could inform land use decisions and policy making processes, notably REDD+. We recommend that deforestation probability based on historical trends are used to stratify country's forest areas to focus future monitoring and the implementation of policies and measures especially in the frame of upcoming REDD+ efforts on those areas with increased probability of deforestation. This information could be useful to policy makers in setting priorities and by overlaying these maps with the world database on protected areas one can identify conservation areas which might receive higher deforestation pressures. This information could be helpful to policy makers in their decisions of conservation priority setting or in the degazetting of protected areas.
The authors would like to thank Kenneth MacDicken for his guidance and for reviewing the article. We are grateful for the financial support from the European Commission and from the SFM in a Changing Climate Programme funded by the Government of Finland. Furthermore, we thank Robert Gilmore Pontius Jr. and Clark University GIS Land Change Science seminar for processing data. Lastly we would like to thank three anonymous referees for their valuable revision and suggestions.
ALEXANDRATOS, N. and BRUINSM, A.J. 2012. World agriculture towards 2030/2050: the 2012 revision. ESA Working paper No. 12-03. Rome, FAO.
AREKHI, A. 2011. Modeling spatial pattern of deforestation using GIS and logistic regression: A case study of northern Ilam forests, Ilam province, Iran. African Journal of Biotechnology 10(72): 16236-16249
DIMICELI, C.M., CARROLL, M.L., SOHLBERG, R.A., HUANG, C., HANSEN, M.C. and TOWNSHEND, J.R.G. 2011. Annual global automated MODIS Vegetation Continuous Fields (MOD44B) at 250 m spatial resolution for data years beginning day 65, 2000--2010, Collection 5 Percent Tree Cover. University of Maryland, College Park, MD, USA. Downloaded from http://www.landcover. org/data/vcf/
ERNST, C., MAYAUX, P., VERHEGGHEN, A., BODART, C., CHRISTOPHE, M. and DEFOURNY, P. 2012. National forest cover change in Congo Basin: deforestation, reforestation, degradation and regeneration for the years 1990, 2000 and 2005. Global Change Biology 2013.
EVA H., CARBONI S., ACHARD F., STACH N., DURIEUX L., FAURE J-F. and MOLLICONE D. 2010. Monitoring forest areas from continental to territorial levels using a sample of medium spatial resolution satellite imagery. ISPRS J. Photogram. Remote Sens. 65: 191-197.
FAO 2002. Pan-tropical survey of forest cover changes 19802000. FRA Working paper 49s. Food and Agriculture Organization of the United Nations, Rome, 42 pp.
FAO 2007. Gridded livestock of the world 2007. By GRW Wint and TP Robinson. Food and Agriculture Organization of the United Nations, Rome, 131 pp.
FAO 2010. Global Forest Resources Assessment, main report. Food and Agriculture Organization of the United Nations, Rome, 340 pp.
FAO 2012. Global ecological zones for FAO forest reporting: 2010 update. By S Iremonger and A Gerrand. Forest Resources Assessment working paper 179. Food and Agriculture Organization of the United Nations, Rome, 52 pp.
FAO and JRC 2012. Global forest land-use change 19902005. By E.J. Lindquist, R. D'Annunzio, A. Gerrand, K. MacDicken, F. Achard, R. Beuchle, A. Brink, H.D. Eva, P. Mayaux, J. San-Miguel-Ayanz & H-J. Stibig. FAO Forestry Paper No. 169. Food and Agriculture Organization of the United Nations and European Commission Joint Research Centre. Rome, 40 pp.
FAO 2014. Global Administrative Unit Layers (GAUL). Geonetwork. Food and Agriculture Organization of the United Nations, Rome http://www.fao.org/geonetwork/ srv/en/metadata.show?id=12691
FAO 2016. Global Forest Resources Assessment 2015. How are the world's forests changing? Second edition. Food and Agriculture Organization of the United Nations, Rome
GEIST, H.J. and LAMBIN, E.F. 2001. What drives tropical deforestation? LUCC Report Series 4. Louvain La Neuve.
GIBBS, H.K., RUESCH, A.S., ACHARD, F., CLAYTOND, M.K., HOLMGREN, P., RAMANKUTTY, N. and FOLEY, J.A. 2010. Tropical forests were the primary sources of new agricultural land in the 1980s and 1990s. PNAS 107(38): 16732-16737.
GUTIERREZ-VELEZ, V.H., DE FRIES, R.S., PINEDOVASQUEZ, M., URIARTE, M., PADOCH, C., BAETHGEN, W., FERNANDES, K. and LIM, Y. 2011. High-yield oil palm expansion spares land at the expense of forests in the Peruvian Amazon. Environ Res Lett 6(044029): 5.
HANSEN, M.C., POTAPOV, P.V., MOORE, R., HANCHER, M., TURUBANOVA, S.A., TYUKAVINA, A., THAU, D., STEHMAN, S.V., GOETZ, S.J., LOVELAND, T.R., KOMMAREDDY, A., EGOROV, A., CHINI, L., JUSTICE, C.O. and TOWNSHEND, J.R.G. 2013. High-Resolution Global Maps of 21st-Century Forest Cover Change. Science (342): 850-853.
HOSONOUMA, N., HEROLD, M., DE SY, V., DE FRIES, R.S., BROCKHAUS, M., VERCHOT, L., ANGELSEN, A. and ROMIJN, E. 2012. An assessment of deforestation and forest degradation drivers in developing countries. Environ. Res. Lett. 7 (044009): 12pp.
IIASA/FAO 2010. Global Agro-ecological Zones (GAEZ v3.0) : Model Documentation. By Fischer, G., Nachtergaele, F.O., Prieler, S., Teixeira, E., Toth, G., van Velthuizen, H., Verelst, L. et al.. International Institute for Applied System Analysis, Laxenburg, Austria.
IBM 2011. IBM SPSS Statistics for Windows, Version 20.0. International Business Machines Corporation, Armonk, NY.
IRWIN, E. and GEOGHEGAN, J. 2001. Theory, data, methods: developing spatially explicit economic models of land use change. Agriculture, Ecosystems and Environment 85: 7-23.
JOPPA, L., LOARIE, S. and PIMM, S. 2008 On the protection of 'protected areas'. Proc. Natl Acad. Sci. USA 105: 6673-6678.
JOPPA, L.N. and PFAFF, A. 2011 Global protected area impacts. Proc. Royal Soc. B: Biol. Sci. 278(1712): 16331638.
JRC 2008. Travel time to major cities: A global map of Accessibility. Joint Research Centre of the European Commission. http://forobs.jrc.ec.europa.eu/products/gam/
LAMBIN, E.F. and MEYFROIDT, P. 2010. Land use transitions: socio-ecological feedback versus socio-economic change. Land Use Policy 27: 108-118.
MATHER, A.S. 2005. Assessing the world's forests. Global Environmental Change 15: 267-280.
MAYAUX, P., PEKEL, J.-F., DESDEE, B., DONNAY, F., LUPI, A., ACHARD, F., CLERICI, M., BODART, C., BRINK, A., NASI, R. and BELWARD, A. 2013. State and evolution of the African rainforests between 1990 and 2010. Phil Trans R Soc B 368: 20120300.
NEPSTAD, D.C., STICKLER, C.L., SOARES-FILHO, B. and MERRY, F. 2008. Interactions among Amazon land use, forests and climate: Prospects for a near-term forest tipping point. Phil Trans R Soc B 363: 1737-1746.
PBL 2013. BIOME model simulation for the year 2000. The Netherlands Environmental Assessment Agency.
PONTIUS JR, R.G. and BATCHU, K. 2003. Using the relative operating characteristic to quantify certainty in prediction of location of land cover change in India. Transactions in GIS 7(4): 467-484.
PONTIUS JR, R.G. and CHEN, H. 2006. GEOMOD Modeling: Land-Use & Cover Change Modeling, Idrisi 15: The Andes edition. Clark Labs, Worcester, MA, 44pp.
PRENTICE, I.C., CRAMER W., HARRISON, S.P., LEEMANS, R., MONSERUD, R.A. and SOLOMON, A.M. 1992. A global biome model based on plant physiology and dominance, soil properties and climate. Journal of Biogeography 19: 177-134.
ROSA, I.M.D, PURVES, D., SOUZA JR, C. and EWERS R.M. 2013. Predictive Modelling of Contagious Deforestation in the Brazilian Amazon. PLoS ONE 8(10): e77231.
RUDEL, T.K., COOMES, O.T., MORAN, E.F., ACHARD, F., ANGELSEN, A., XU, J. and LAMBIN, E.F. 2005. Forest transitions: towards a global understanding of land use change. Global Environmental Change 15: 23-31.
SALVATORE ET AL., 2005. Mapping Global Urban and Rural Population Distributions. Food and Agriculture Organization of the United Nations, Rome.
SMITH P, GREGORY PJ, VAN VUUREN D, OBERSTEINER M, HAVLIK P, ROUNSEVELL M, WOODS J, ET AL. 2010. Competition for land. Philosophical transactions of the Royal Society of London. Series B, Biological sciences, 365(1554): 2941-57.
SONTER L.J., BARRETT D.J., SOARES-FILHO B.S. and MORAN C.J. 2014. Global demand for steel drives extensive land-use change in Brazil's Iron Quadrangle. Global Environmental Change 26: 63-72.
TREWARTHA, G.T. 1968. An Introduction to Climate, 4th edition. McGraw-Hill, New York.
WASSENAAR, T., GERBERA, P., VERBURG, P.H., ROSALES, M. IBRAHIM, M. and STEINFELD, H. 2007. Projecting land use changes in the Neotropics: The geography of pasture expansion into forest. Global Environmental Change 17(2007): 86-104.
WILLIAMS, M. 2000. Dark ages and dark areas: global deforestation in the deep past. Journal of Historical Geography 26(1): 28-46.
WILLIAMS, M. 2006. Deforesting the Earth: From Prehistory to Global Crisis. Reviewed by William Meyer. Annals of the Association of American Geographers 96: 674-675.
ZON, R. and SPARHAWK, W.W. 1923. Forest resources of the world. McGraw-Hill, New York, 997 pp.
M. SANDKER (a), Y. FINEGOLD (a,b), R. D'ANNUNZIO (a) and E. LINDQUIST (a)
(a) Food and Agriculture Organization of the United Nations, Forestry Policy and Resources Division, Forestry Department, Rome, Italy
(b) Clark University, Worchester, Massachusetts, USA
Email: Marieke.Sandker@fao.org, Yelena.Finegold@fao.org, Remi.DAnnunzio@fao.org, Erik.Lindquist@fao.org
Caption: FIGURE 1 Approximation of historical (pre-agriculture) forest cover and historical net forest loss (approximately between 4000 B.C.-2000 A.D.). Loss as a result of climate change over this extensive period is not included. Resolution: 10x10km
Caption: FIGURE 2 Schematic overview of how the graphical display of the forest/non-forest ratio distribution per variable class is obtained
Caption: FIGURE 3 Historical (a) and recent (b) forest loss plotted against accessibility (cost-distance). E.g. 70% of the forest at a costdistance of 30 minutes or less has been permanently lost over the past 6000 years worldwide
Caption: FIGURE 4 Historical (a) and recent (b) forest loss plotted against rural population density
Caption: FIGURE 5 Historical (a) and recent (b) forest loss plotted against ruminant livestock density. The input map used did not provide more detail which is why the classes could not be further disaggregated
Caption: FIGURE 6 Historical (a) and recent (b) forest loss plotted against crop suitability. The crop suitability index (SI) is defined as attainable yields as a percentage of the maximum constraint-free yield, the maximum SI therefore being 100
Caption: FIGURE 7 Historical (a) and recent (b) forest loss plotted against pasture suitability. Average attainable output densities are in tons/ha of grass/pasture yields
Caption: FIGURE 8 Economic development and recent forest loss (1990-2005) in the tropics: the forest transition curve
Caption: FIGURE 9 Forest loss probability map with regional boundaries based on the correlation between historical forest loss with accessibility, rural population density and crop suitability. Resolution: 10x10km
TABLE 1 Variables for which a correlation with deforestation was explored Direct cause it is Variable Description expected to influence Rural Rural population density Agricultural expansion population is represented in persons (crop and pasture); density per km2. Rural areas are clearing for wood; defined as "what is not infrastructure expansion; urban" and vary by localized and national definitions of urban. Accessibility Travel time to the Agricultural expansion; nearest city of 50,000 infrastructure expansion; people or more, where clearing for wood; urban travel time is calculated expansion; mining including information on slope, elevation, roads, railways, navigable rivers, major waterbodies and national borders. Crop The generalized crop Cropland expansion suitability suitability map is an assembly of attainable yield levels for rain-fed crop production of 19 individual crop and crop group suitability maps as produced by the Global Agro-Ecological Zones (GAEZ). GAEZ combines data on climate, soil and terrain compiled to agronomically meaningful resource inventories. Pasture Average attainable output Pastureland expansion suitability density for grass and pasture yields assuming low input levels as produced by GAEZ and expressed in tons/ha. Ruminant Expressed as Tropical Pastureland expansion livestock Livestock Units (with density equivalent live-weight of 250 kg) in the year 2000. Variable Source and resolution Rural Salvatore et al. 2005; population 10[km.sup.2] density Accessibility JRC 2008; 30 arc second, resampled to 250m Crop GAEZ (IIASA/FAO 2010); 30 suitability arc seconds Pasture GAEZ 2010; original suitability resolution = 30 arc seconds Ruminant FAO 2007 in GAEZ 2010; livestock original resolution= 30 density arc seconds TABLE 2 Climatic domains, ecozones and their criteria, modified from FAO (2012) Climatic domain Ecozone Climatic and orographic criteria Tropical Tropical rain forest Wet: 0-3 months dry Tropical moist forest Wet/dry: 3-5 months dry Tropical dry forest Dry/wet: 5-8 months dry Tropical shrubland Semi-Arid: Evaporation > Precipitation Tropical desert Arid: All months dry Tropical mountain Approximate > 1 000 m systems altitude (local variations) Subtropical Subtropical humid forest Humid: No dry season Subtropical dry forest Seasonally Dry Subtropical steppe Semi-Arid: Evaporation > Precipitation Subtropical desert Arid: All months dry Subtropical mountain Approximate > 800-1 000 systems m altitude Temperate Temperate oceanic forest Oceanic climate: coldest month over 0[degrees]C Temperate continental Continental climate: forest coldest month under 0[degrees]C Temperate steppe Semi-Arid: Evaporation > Precipitation Temperate desert Arid: All months dry Temperate mountain Approximate > 800 m systems altitude Boreal Boreal coniferous forest Vegetation physiognomy: coniferous dense forest dominant Boreal tundra woodland Vegetation physiognomy: woodland and sparse forest dominant Boreal mountain systems Approximate > 600 m altitude Polar Same as domain level TABLE 3 Strata used to allocate forest quantities as assessed by FAO & JRC 2012 for the creation of the 2000 forest map Stratum VCF threshold (lowest VCF value considered forest in the stratum) Tropical Africa 20 Subtropical Africa 22 Tropical Asia 40 Subtropical Asia 28 Temperate Asia 25 Boreal Asia 17 Subtropical Europe 21 Temperate Europe 21 Boreal Europe 13 Tropical North America 33 Subtropical North America 23 Temperate North America 28 Boreal North America 14 Tropical Oceania 13 Subtropical Oceania 23 Temperate Oceania 20 Tropical South America 28 Subtropical South America 25 Temperate South America 28 TABLE 4 Economic characterization (authors' approximation) of the different phases of forest transition Transition Economic characterization Source for GDP/GDP phase change Pre Low income, low increase Gross World Bank 2013 Domestic Product (GDP) <1 000 USD- capita in 2000 <30% increase GDP between 1990-2005 Early Low income, high increase GDP <1 000 USD/capita in 2000 >30% increase GDP between 1990-2005 Late Middle income GDP 1 000-10 000 USD/capita in 2000 Post High income >10 000 USD/capita in 2000 TABLE 5 The approximation of historical (pre-agriculture) and recent (1990-2005) forest loss Ecological domain Approximated historical Recent gross forest forest loss (%) (1) loss (%) (1) Tropical 43-53 (2) 7.1 Subtropical 54-60 (2) 4.0 Temperate 48 1.6 Boreal 19 1.2 All domains 39-46 (2) 5.4 (1) Historical loss is expressed as a percentage of total potential forest in each domain. Recent gross forest loss is expressed as percentage of the forest area in 1990 in each domain. (2) The lower estimates exclude the dry forest zone, highest estimates include dry forest zone as proxy of pre-agriculture forest cover. TABLE 6 Results of ROC analysis: strength of relationship between historical deforestation and the variables. An AUC value of 0.5 represents a perfectly random relationship, while a value of 1 means a perfectly strong relationship Historical deforestation (AUC) Accessibility 0.68 * Rural population density 0.73 Crop suitability 0.65 Pasture suitability 0.53 * negative relationship, AUC is 1-AUC. TABLE 7 Tropical forest land (2000) by accessibility and crop suitability class Accessibility Crop suitability 0-2 hours 2-6 hours 6-12 hours very high 0.1% 0.6% 0.5% good 1% 5% 6% moderate 4% 11% 11% marginal 2% 5% 5% not suitable 0.01% 0.01% 0.02% % of total forest in tropics 7% 23% 23% Accessibility Crop suitability 12-24 hours >24 hours very high 0.2% 0.05% good 5% 5% moderate 11% 14% marginal 6% 8% not suitable 0.01% 0.005% % of total forest in tropics 21% 27% Accessibility % of total forest Crop suitability in tropics very high 1% good 23% moderate 50% marginal 25% not suitable 0.1% % of total forest in tropics 100% TABLE 8 Correlation matrix of driver variables (n= 5 845) Kendall's tau b correlation coefficients Rural population Ruminant density livestock density Kendall's Rural population 1.000 .365 tau b density Ruminant livestock 1.000 density Pasture suitability Accessibility Crop suitability Kendall's tau b correlation coefficients Pasture Accessibility suitability Kendall's Rural population .170 -.354 tau b density Ruminant livestock .155 -.329 density Pasture suitability 1.000 -.228 Accessibility 1.000 Crop suitability Kendall's tau b correlation coefficients Crop suitability Kendall's Rural population .147 tau b density Ruminant livestock .144 density Pasture suitability .311 Accessibility -.231 Crop suitability 1.000 TABLE 9 Pressure on forests within protected areas (this does not directly reflect the actual expectation of deforestation but may be indicative for the effort of avoiding deforestation in these protected areas). Read as: 17 % of the forest in Southern Africa (1.3 mln ha) is under protection, of this 1.3 mln ha 12% is at low risk and 88% at high risk of being deforested Southern Africa Western Africa 2010 forest 7.90 46 area (mln ha) of which protected 1.3 (17%) 2.8 (6%) in 2010 (mln ha) non- protected non- protected protected protected % forest at low 4% 12% 8% 40% risk % forest at medium- 96% 88% 92% 60% high risk Eastern Africa Central Africa 2010 forest 160 345 area (mln ha) of which protected 22 (14%) 24 (7%) in 2010 (mln ha) non- protected non- protected protected protected % forest at low 25% 58% 62% 84% risk % forest at medium- 75% 42% 38% 16% high risk Northern Africa 2010 forest 17 area (mln ha) of which protected 1.2 (7%) in 2010 (mln ha) non- protected protected % forest at low 4% 64% risk % forest at medium- 96% 36% high risk * Medium-high risk is defined by including 50% of the suitability for loss values
|Printer friendly Cite/link Email Feedback|
|Author:||Sandker, M.; Finegold, Y.; D'annunzio, R.; Lindquist, E.|
|Publication:||International Forestry Review|
|Date:||Sep 1, 2017|
|Previous Article:||Learning landscape sustainability and development links/Apprendre les liens entre la durabilite et le developpement des paysages/Aprendizaje sobre la...|
|Next Article:||Reduced impact logging and silvicultural interventions in Ghana: the case of Bobiri Forest Reserve/Coupe de bois a impact reduit et interventions...|